Leveraging spectral diversity for machine learning-based estimation of radio frequency signal parameters

Abstract

An example method for estimating the angle-of-arrival (AoA) and other parameters of radio frequency (RF) signals that are received by an antenna array comprises: receiving a plurality of radio frequency (RF) signal power measurements by a plurality of antenna elements at a plurality of RF channels; computing, by applying a machine learning model to the plurality of RF signal power measurements, an estimated RF signal parameter value; and outputting the RF signal parameter value.

What is claimed is: 
     
       1. A method, comprising:
 receiving, by a processing device, a first indication that a first estimated error value does not satisfy a first predetermined threshold value, the first estimated error value related to (i) a first number of signal paths and (ii) a first estimated characteristic of a radio frequency (RF) signal observed by an antenna array; 
 computing, by the processing device, responsive to receiving the first indication that the first estimated error value does not satisfy the first predetermined threshold value, a second estimated error value related to (i) a second number of signal paths larger than the first number of signal paths and (ii) a second estimated characteristic of the RF signal; 
 receiving, by the processing device, a second indication that the second estimated error value satisfies a second predetermined threshold value; and 
 outputting, by the processing device, the second estimated characteristic of the RF signal. 
 
     
     
       2. The method of  claim 1 , further comprising computing, by the processing device, the first estimated error value, wherein (i) the first estimated error value is computed using a first machine learning model trained with a first training data set produced in an environment comprising the first number of signal paths and (ii) the second estimated error value is computed using a second machine learning model trained with a second training data set produced in an environment comprising the second number of signal paths. 
     
     
       3. The method of  claim 2 , wherein use of the first machine learning model requires less processing than use of the second machine learning model. 
     
     
       4. The method of  claim 2 , further comprising:
 before computing the first estimated error value:
 computing, by the processing device, a third estimated error value related to (i) a third number of signal paths smaller than the first number of signal paths and (ii) a third estimated characteristic of the RF signal; and 
 receiving, by the processing device, a third indication that the third estimated error value does not satisfy a third predetermined threshold value. 
 
 
     
     
       5. The method of  claim 4 , wherein the third estimated error value is computed using a third machine learning model is trained with a third training set produced in an environment comprising the third number of signal paths. 
     
     
       6. The method of  claim 2 , wherein use of the first machine learning model requires less processing than use of the second machine learning model. 
     
     
       7. The method of  claim 1 , wherein the second estimated characteristic of an RF signal is one of a line-of-sight (LoS) angle-of-arrival (AoA), a reflection AoA, an attenuation of reflection, a relative delay of reflection, or a relative phase. 
     
     
       8. The method of  claim 1 , wherein the first number of signals is one less than the second number of signals. 
     
     
       9. A system comprising:
 a transceiver configured to receive a plurality of radio frequency (RF) signal power measurements observed by an antenna array; and 
 a processor coupled to the transceiver, the processor to:
 receive a first indication that a first estimated error value does not satisfy a first predetermined threshold value, the first estimated error value related to (i) a first number of signal paths and (ii) a first estimated characteristic of the plurality of RF signal power measurements observed by the antenna array; 
 compute, responsive to receiving the first indication that the first estimated error value does not satisfy the first predetermined threshold value, a second estimated error value related to (i) a second number of signal paths larger than the first number of signal paths and (ii) a second estimated characteristic of the plurality of RF signal power measurements; 
 receive a second indication that the second estimated error value satisfies a second predetermined threshold value; and 
 output the second estimated characteristic of the plurality of RF signal power measurements. 
 
 
     
     
       10. The system of  claim 9 , wherein the processor is further to compute the first estimated error value, wherein (i) the first estimated error value is computed using a first machine learning model trained with a first training data set produced in an environment comprising the first number of signal paths and (ii) the second estimated error value is computed using a second machine learning model trained with a second training data set produced in an environment comprising the second number of signal paths. 
     
     
       11. The system of  claim 10 , wherein use of the first machine learning model requires less processing than use of the second machine learning model. 
     
     
       12. The system of  claim 10 , further comprising:
 before computing the first estimated error value:
 compute a third estimated error value related to (i) a third number of signal paths smaller than the first number of signal paths and (ii) a third estimated characteristic of the plurality of RF signal power measurements; and 
 receive a third indication that the third estimated error value does not satisfy a third predetermined threshold value. 
 
 
     
     
       13. The system of  claim 12 , wherein the third estimated error value is computed using a third machine learning modem trained with a third training data set produced in an environment comprising the third number of signal paths. 
     
     
       14. The system of  claim 10 , wherein use of the first machine learning model requires less processing than use of the second machine learning model. 
     
     
       15. The system of  claim 9 , wherein the second estimated characteristic of the plurality of RF signal power measurements is one of a line-of-sight (LoS) angle-of-arrival (AoA), a reflection AoA, an attenuation of reflection, a relative delay of reflection, or a relative phase. 
     
     
       16. The system of  claim 9 , wherein the first number of signals is one less than the second number of signals. 
     
     
       17. A method comprising:
 training a first machine learning model using a first dataset produced in a first signal propagation environment, the first signal propagation environment having a first number of signal paths measurable by an antenna array; 
 training a second machine learning model using a second dataset produced in a second signal propagation environment, the second signal propagation environment having a second number of signal paths measurable by an antenna array, wherein the second number of signal paths is larger than the first number of signal paths, wherein using the first machine learning model requires less processing than using the second machine learning model. 
 
     
     
       18. The method of  claim 17 , wherein the first signal propagation environment has a direct signal path and no reflected signal paths and the second signal propagation environment has a direct signal path and at least one reflected signal path. 
     
     
       19. The method of  claim 17 , further comprising:
 training a third machine learning model using a third dataset produced in a third signal propagation environment, the third signal propagation environment having a third number of signal paths measurable by an antenna array, wherein the third number of signal paths is larger than the second number of signal paths, wherein using the second machine learning model requires less processing than using the third machine learning model. 
 
     
     
       20. The method of  claim 19 , wherein the third signal propagation environment has a direct signal path and at least one reflected signal path more than the second signal propagation environment.